The Lumer Phillips Theorem The Hille-Yosida theorem provides a set of conditions on the operator A that are necessary and sufficient in order for -A to generate a strongly continuous semigroup of contractions. It is often more convenient to have an equivalent set of conditions as described in the following theorem. Theorem (Lumer-Phillips) Suppose A : D A H is closed and densely defined. Then the following are equivalent conditions: −A generates a C 0 semigroup of contractions on H A is accretive and I + A : D A H is surjective 1. 2. Proof- 2 1: Suppose A is accretive. Then for λ > 0 ||λI + Ax|| H ≥ λ||x|| H ∀x ∈ D A Letting z = λI + Ax this becomes, ||z|| H ≥ λ||λI + A −1 z|| H . In particular, since I + A is surjective, this previous estimate implies ||I + A −1 || LH ≤ 1. Now observe that if 0 < λ < 2, then formally, λI + A −1 = I + A −1 I + λ − 1I + A −1 −1 ∞ = I + A −1 ∑ n=0 1 − λI + A −1 n Note that if 0 < λ < 2, then 1 − λI + A −1 ∈ LH and ||1 − λI + A −1 || LH ≤ |λ − 1|≤ 1. Then the Neumann series for 1 − λI + A −1 converges and we get, I + λ − 1I + A −1 −1 ∞ = ∑ n=0 1 − λI + A −1 n Now let B = I + λ − 1I + A −1 and note that λ + AI + A −1 B −1 = λ − 1 + I + AI + A −1 B −1 = λ − 1I + A −1 + IB −1 = BB −1 = I and I + A −1 B −1 λ + A = I + A −1 B −1 λ − 1 + I + A = I + A −1 B −1 λ − 1I + A −1 + II + A = I + A −1 B −1 BI + A = I; i.e., λ + A −1 = I + A −1 B −1 and 1 ||λ + A −1 || LH = ||I + A −1 || LH ||B −1 || LH ∞ ≤ ∑ n=0 1 − λ n = 1/λ 0 < λ < 2. Now it follows that μI + A −1 = λI + A −1 I + μ − λλI + A −1 −1 and for μ > 0 and |μ − λ|||λI + A −1 || LH < 1, we have μI + A −1 exists and ||μI + A −1 || LH ≤ 1/μ Repeating this process we see that λI + A −1 exists for all λ > 0, and ||λI + A −1 || LH ≤ 1/λ Then the hypotheses of the Hille-Yosida theorem are satisfied and −A generates a C 0 semigroup of contractions on H. This proves (2) implies (1). Proof that 1 2 : If −A generates a C 0 semigroup of contractions on H then Stx − x, x H = Stx, x H − ||x|| 2H ≤ ||Stx|| H − ||x|| H ||x|| H ≤ 0 This implies that for all x in D A Stx − x/t, x H −Ax, x H ≤ 0 i.e., A is accretive. In addition, it follows from the Hille-Yosida theorem that λ + A is surjective for all λ > 0, and for λ = 1 in particular.■ Examples1) Let H = H 0 R , A = ∂ x , D A = H 1 R. Note that u ∈ H 1 R implies that ux is continuous and tends to zero as |x| tends to infinity. Then Au, u H = ∫ u ′ xuxdx = 1/2 ux 2 | x=∞ x=−∞ = 0 R ∀u ∈ D A and A is accretive. Note that according to this, − A is also accretive. Next, note that I ± Au = ux ± u ′ x = fx u ∈ DA, f ∈ H implies 1 ± iαUα = Fα where U, F denote Fourier transforms of u and f, respectively. Then Uα = Fα Fα = 1 ∓ iα = Gα ∓ iαGα 1 ± iα 1 + α2 and ux = gx ∓ g ′ x where gx = T −1 F Fα 1 + α2 = ∫ e |x−y| fydy. R This shows that I + A : D A H is onto and now it follows from the Lumer-Phillips theorem that −A generates S(t), a C 0 − s/g of contractions, on H. This implies, in turn that for each 2 u 0 ∈ D A there exists a unique ut satisfying, u′t + Aut = ∂ t ux, t + ∂ x ux, t = 0, u0 = u 0 . Since it is evident that ux, t = u 0 x − t solves the initial value problem, it follows by uniqueness that Stu 0 x = u 0 x − t. Since −A is also accretive and I − A : D H is onto, +A must also generate a C 0 − s/g of contractions, Zt. Then the solution of u′t − Aut = ∂ t ux, t − ∂ x ux, t = 0, u0 = u 0 ∈ D A is given by ux, t = Ztu 0 x = u 0 x + t = S−tu 0 x. Then uniqueness implies that Zt = S−t. In this case, the operator A generates a C 0 group of contractions on H. That is, Ut = if t≥0 Z−t if t<0 St satisfies UtU−t = StS−t = S0 = I. Then Ut : t ∈ R is a group of contractions on H since for each operator in the collection, the inverse operator is also in the collection. 2) Let H = H 0 U , for U ⊂ R n , open and bounded. Let Au = −divM∇u, for M a symmetric positive definite matrix, and let D A = H 10 U ∩ H 2 U. Then Au, u H = Bu, u = ∫ ∇u ⋅ M∇udx ≥ 0 U ∀u ∈ D A . This shows that A is accretive and, from previous results regarding elliptic operators, it is known that A + λI is surjective for λ ≥ 0. For λ = 1, in particular, A + I is surjective and the Lumer-Phillips theorem asserts that −A generates a C 0 − s/g of contractions, S(t). Then the IVP u ′ t + Aut = ft, u0 = u 0 has a unique solution for all u 0 ∈ D A and for all f ∈ C 1 0. ∞ : H. The solution is given by t ut = Stu 0 + ∫ St − τfτdτ. 0 This abstract IVP stands for the problem ∂ t ux, t − divM∇ux, t = fx, t ux, 0 = u 0 x ux, t = 0 in U T in U on Γ × 0, T and then t ux, t = ∑ n>0 e −λ n t u 0 , φ n H φ n x + ∑ n>0 ∫ e −λ n t−τ f⋅, τ, φ n H dτ φ n x, 0 where λ n , φ n denote the eigenvalues and normalized eigenfunctions associated with the elliptic operator A. Evidently Stu 0 = ∑ n>0 e −λ n t u 0 , φ n H φ n x. 3 Groups A family Gt : t ∈ R ⊂ LH is said to form a group on H if: i ii GtGs = Gt + s G0 = I ∀s, t ∈ R Note that GtG−t = G0 = I so that for each real t, Gt −1 = G−t. The group G(t) is said to be a C 0 -group on H if, Gtx x in H as t tends to 0, for all x in H. As in the case of semigroups, there is a closed, densely defined linear operator associated with G(t), Bx = lim h0 Gh − I x h for all x ∈ D B = x ∈ H : lim h0 Gh − I x exists h and we say this operator B is the generator for the group. Note that if G(t) is generated by B, then B generates a C 0 -s/g, St = Gt for t ≥ 0, and −B generates a C 0 -s/g, S−t = Gt for t ≤ 0. In this case, the Lumer-Phillips theorem implies that B is accretive and I + B : D B H is surjective and, in addition, − B is accretive and I − B : D B H is surjective. Note that when B and −B are both accretive, then Bx, x H ≥ 0 and −Bx, x H ≥ 0 Bx, x H = 0 ∀x ∈ D B ∀x ∈ D B i.e., In this case we say the operator B is conservative. Note that this implies d/dt||Gtx|| 2H = 2G ′ tx, Gtx H = 2BGtx, Gtx H = 0 i.e., ||Gtx|| 2H = ||x|| 2H for all t ∈ R Then ||Gt|| LH = 1 and we say that Gt : t ∈ R is a unitary group on H. Example- Consider the IVP ∂ tt ux, t − ∂ xx ux, t = 0, ux, 0 = fx, ∂ t ux, 0 = gx u0, t = u1, t = 0, u1 = ∂xu u2 = ∂tu Let 0 < x < 1, t > 0 t > 0, 0 < x < 1. ∂ t u 1 = ∂ xt u = ∂ x u 2 ∂ t u 2 = ∂ tt u = ∂ x u 1 Then ∂t u1 u2 − 0 1 1 0 ∂x u1 u2 = 0 0 and u1 u2 t = 0 = f ′ x gx i.e., 4 ∂ t Ut − AUt = 0, U0 = U 0 where H = L 2 0, 1 2 , A= 0 1 1 0 ∂x D A = U ∈ H : u 1 ∈ H 1 0, 1, u 2 ∈ H 10 0, 1 Then 1 1 0 0 AU, U H = ∫ ∂ x u 2 ⋅ u 1 + u 2 ⋅ ∂ x u 1 dx = ∫ d/dxu 1 u 2 dx 1 = u 1 u 2 | x=1 x=0 = 0 (since u 2 ∈ H 0 0, 1 This proves A is conservative. Now for λ ≠ 0, F ∈ H, consider λ u1 u2 +A u1 u2 = λu 1 + ∂ x u 2 λu 2 + ∂ x u 1 = F1 F2 Then λ∂ x u 1 + ∂ xx u 2 = ∂ x F 1 and ∂ x u 1 = F 2 − λu 2 , or ∂ xx u 2 − λu 2 = ∂ x F 1 − λF 2 Since ∂ x F 1 − λF 2 ∈ H −1 0, 1, this last equation has a unique weak solution u 2 ∈ H 10 0, 1, by the previously developed elliptic theory. Then λu 1 = F 1 − ∂ x u 2 ∈ L 2 0, 1, ∂ x u 1 = F 2 − λu 2 ∈ L 2 0, 1, so u 1 ∈ H 1 0, 1 and U ∈ D A . This shows that λ + A : D A H is surjective for all λ ≠ 0. ■ We have the following version of the Hille-Yosida theorem for groups rather than semigroups. Theorem The following statements are equivalent: 1. 2. B : D B H generates a C 0 − unitary group on H a) B is closed and densely defined b) ∀λ ≠ 0 λ − B : D A H is one to one and onto with λ − B −1 LH ≤ 1 λ Proof- Suppose B generates a group, {Gt : t ∈ R. Then B generates a contraction semigroup, Gt : t ≥ 0 and −B also generates a contraction semigroup, G−t : t ≥ 0. Then Hille-Yosida theorem implies that both B and − B satisfy the necessary conditions for generating a contraction semigroup and this implies 2. Conversely, suppose the conditions 2 hold. Then, again by the Hille-Yosida theorem, B generates a contraction semigroup S + t : t ≥ 0 and − B generates a contraction semigroup S − t : t ≥ 0, and these semigroups commute. For x 0 ∈ D B 5 d dt S + tS − tx 0 = 0 t≥0 from which it follows that S + tS − t = I t ≥ 0. Then Gt = S + t t≥0 S − −t t≤0 satisfies all of the conditions for a group. i.e., (i) and (ii) are evident and 1 = ||Gt G−t|| ≤ ||Gt|| ||G−t|| ≤ ||Gt|| ≤ 1. Finally, to see that B generates G(t), note that B generates S + t : t ≥ 0 so for h > 0, S + h − Ix Ghx − x = Bx. h h and − B generates S − t : t ≥ 0, so for h < 0 Ghx − x S −h − Ix =− − −−Bx. ■ h −h We have also a version of the Lumer-Phillips theorem for groups. Theorem The following statements are equivalent: 1. 2. . B : D B H generates a C 0 − unitary group on H a) B is closed and densely defined b) B is conservative c) λ + B is onto for some λ > 0 and for some λ < 0. Example- Consider the IVP for the wave equation ∂ tt ux, t − ∂ xx ux, t = 0 ux, 0 = fx, ∂ t ux, 0 = gx, u0, t = u1, t = 0, Let 0 < x < 1, 0 < t < T, 0 < x < 1, 0 < t < T. u 1 x, t = ∂ x ux, t u 2 x, t = ∂ t ux, t so ∂ t u 1 x, t = ∂ xt ux, t = ∂ x u 2 x, t ∂ t u 2 x, t = ∂ tx ux, t = ∂ x u 1 x, t u 1 x, 0 = f ′ x u 2 x, 0 = gx; i.e., ⃗ t = ∂tU 0 1 1 0 ⃗ t, ∂xU ⃗ x, 0 = U f′ g . Let H = L 2 0, 1 2 , DA = ⃗ ∈ H : u 1 ∈ H 1 0, 1, u 2 ∈ H 10 0, 1 U Then ⃗, U ⃗ AU 1 H 1 d dx = ∫ u 1 ∂ x u 2 + u 2 ∂ x u 1 dx = ∫ 0 0 u 1 u 2 dx = u 1 u 2 | x=1 x=0 = 0 which shows that A is conservative and must therefore generate a group if we can show 6 ⃗ = that condition 2c of the previous theorem is satisfied. But for λ ≠ 0 and F ⃗ + AU ⃗ =F ⃗ λU λu 1 + ∂ x u 2 λu 2 + ∂ x u 1 = F1 F2 ∈H F1 F2 λ∂ x u 1 + ∂ xx u 2 = ∂ x F 1 But in this case, ∂ x u 1 = F 2 − λu 2 and hence λF 2 − λu 2 + ∂ xx u 2 = ∂ x F 1 , or ∂ xx u 2 − λ 2 u 2 = ∂ x F 1 − λF 1 . This last equation is uniquely solvable with u 2 ∈ H 10 0, 1 since ∂ x F 1 − λF 1 ∈ H −1 0, 1. Next, λu 1 + ∂ x u 2 = F 1 implies λu 1 = ∂ x F 1 − ∂ x u 2 ∈ L 2 0, 1. In addition, ∂ x u 1 = F 2 − λu 2 ∈ L 2 0, 1 hence u 1 ∈ H 1 0, 1 and ⃗ = U u1 u2 ∈ H 1 0, 1 H 10 0, 1 = DA. This shows that λ + A : D A H is onto for λ ≠ 0. Then A generates a group which we know from previous experience is given by ⃗ x, 0 = GtU 1 2 1 2 f̃ ′ x + t + ̃f ′ x − t + f̃ ′ x + t − ̃f ′ x − t + 1 2 1 2 g̃ x + t − g̃ x − t g̃ x + t + g̃ x − t where ̃f, g̃ denote the odd 2-periodic extensions of f and g. 7